I just got back from Snowflake Summit, and honestly, the most interesting part of the experience wasn’t the keynotes. It was the side conversations. The customer meetings. The “this is what’s actually breaking inside my company right now” moments.
Because if you listened closely, there was a pattern: everyone is talking about AI. But what everyone I spoke to is really dealing with… is everything that comes before AI.
Everyone Says They Want AI. Most Teams Are Still Stuck on Dashboards.
Teams did what they were supposed to do over the last couple of years. They moved to Snowflake, ingested data, built pipelines, created dashboards, proved “value.” And now they’re stuck.
One customer told me they’ve got the majority of their operational data centralized, pipelines running cleanly—and yet the team is still building dashboards because that’s what the business keeps asking for. At the same time, leadership is asking a completely different question: “Why don’t we have AI yet?”
That tension is very real. Because on one hand the people building dashboards know something leadership doesn’t always see—those dashboards are often the only thing holding consistency together right now. If you remove them too early, things break. On the other hand, change can be uncomfortable and especially hard to make when you have your day-to-day job to keep up with and the technology changes so fast, nonstop.
So, we’re in this awkward middle phase where everyone knows dashboards aren’t the future . . . but no one can fully let them go.
The Dirty Secret: Most Companies Don’t Trust Their Own Data Enough for AI
The moment conversations moved from reporting to AI, everything slowed down.
- “What does this KPI actually mean?”
- “Why is it different across systems?”
- “Where is the logic defined?”
And the answers were . . . messy.
I heard logic living in BI tools instead of Snowflake, creating both performance issues and inconsistencies, or teams that couldn’t tell which fields were actually used in reports or what data was even safe to deprecate.
In another case, the conversation wasn’t even about AI yet—it was about whether the organization could confidently define and standardize metrics across teams.
And this is exactly the problem no one wants to admit: AI doesn’t fail because of the model. It fails because the business hasn’t agreed on what the data means.
This is Where Things Get Real: Semantics, Ontologies . . . And The Stuff Nobody Wants to Own
Before, “semantic layer” was a BI term. Now it’s turning into the most important—and most uncomfortable—part of the conversation.
Because what customers are running into isn’t a data structure problem. It’s a meaning problem.
One team put it clearly: if you want an AI agent to give consistent answers, you need defined semantics it can follow—otherwise it will just generate outputs based on incomplete logic.
This is where the conversation gets uncomfortable. It’s no longer about pipelines. It’s about ontologies, knowledge graphs, context layers—shared definitions of what your business concepts actually mean and how they relate to each other—the stuff nobody wants to own, but everyone suddenly needs.
Not because people want to build them . . . but because they’re realizing they don’t have a choice.
If AI is going to answer questions, trigger actions, and automate workflows, it needs to understand how your business operates—not just what tables exist.
One conversation really stuck with me. The shift isn’t just about connecting data anymore. It’s about connecting data to process—how decisions are actually made inside the organization.
And most companies? They don’t have that layer built. Or worse: it exists across five tools, three teams, and one person’s brain.
Meanwhile, The BI Layer is Slowly Fading Into the Background
No one is ripping out BI tools tomorrow. But they’re definitely questioning it.
One customer said it out loud: if an agent can answer business questions directly, what role does BI even play long-term?
Others are dealing with the more painful version: logic locked in DAX, data extracted and disconnected from Snowflake, layers of transformations no one fully understands.
The companies that see this coming aren’t waiting for BI vendors to solve it—they’re pulling logic back into the data layer and making the front end replaceable.
One team described their goal as making their data completely front-end agnostic. Which is where this is all going.
This Is Exactly Where Snowflake is Steering at Summit 2026
If you step back and look at the announcements, the direction makes sense. Messaging this year is centered on agentic AI, CoWork (Snowflake Intelligence), CoCo (Cortex Code), and a hard push on context, governance, and identity.
Snowflake isn’t a data platform anymore. It moved from “where data lives” to “where decisions actually happen.” And now to “where those decisions can be executed.”
The Problem Is Most Companies Aren’t Ready to Meet Snowflake There Yet
The same blockers showed up in nearly every conversation: no time to prototype, limited teams, tech debt from earlier phases, logic scattered across tools, and unclear ownership of definitions.
One comment I won’t forget: a customer told me they don’t need a partner to just build what they ask for — they need a data therapist, someone who helps them figure out how to do it better. That gap — between what’s possible and what actually gets delivered — is where most companies are stuck.
My Takeaway (The Honest Version)
We’re not in the “AI transformation” phase yet. We’re in the phase of cleaning up definitions, standardizing logic, and making data usable before AI can touch it.
The companies that take this seriously — that invest in semantics, start building real context layers (whether you call it an ontology or not), and bring logic back into the data platform — will move fast.
Everyone else will stay where they are: building dashboards… and calling it AI progress.
The Part No One Says On Stage
Agentic AI isn’t going to fail because the tech isn’t ready. It’s going to fail because most companies are trying to automate decisions they don’t fully understand or agree on.
You can’t scale AI if every team defines metrics differently, logic lives in different tools, context isn’t documented, and governance is an afterthought. All AI does is amplify what’s already there. Messy foundation, messy output — just faster.
Where Hakkōda, an IBM Company, Fits In
This is also where I’m seeing the role of SI partners change in real time, and it’s a bigger shift than most people are acknowledging.
It’s not “can you implement Snowflake” anymore. Everyone can do that. The real ask—whether clients say it directly or not—is harder: help me understand what my data actually means, untangle the logic I’ve scattered across five tools, and build something that won’t collapse the moment AI touches it.
And the pressure is real. Every leader right now has an AI goal sitting above their head. It’s on the roadmap, it’s in the board deck, it’s in the performance review. But you can’t hit that goal by skipping the foundation work. The fastest path to AI that delivers is through the messy, unglamorous work that nobody wants to put on a slide.
That’s where I’m spending most of my time right now. Not delivering pipelines. Not building dashboards. But sitting in that uncomfortable middle—between data, business logic, and the context AI needs to function. Helping clients figure out what they don’t know yet, not just execute what they’ve already decided.
It’s messier than a demo. It’s slower than a sprint. It’s also where the real work is—and where the real value gets created.
Working Through the Alignment Bottleneck
At Summit, everything was about autonomy. Agents. Automation. But in real conversations, most teams are still trying to answer a much simpler question: “Why is this number different in two reports?”
Most teams think they need better AI. What they need is better alignment. And until that’s solved, AI isn’t transformation. It’s just faster confusion.
Want to learn how Hakkoda and IBM can help your enterprise find the alignment it needs to succeed with AI? Let’s talk today.
